Customer Data Platform (CDP)

GLOSSARY

Definition

A customer data platform collects and unifies customer data from every touchpoint, website visits, email opens, purchases, support tickets, into a single profile that other systems can query. Unlike a CRM, which sales teams use to track deals, a CDP is built for marketers who need a real-time, always-on view of each customer across channels.

Why it matters

CDPs became a distinct category around 2016 when the CDP Institute formed. Before that, marketers stitched together data from a CRM, an email platform, and a web analytics tool, and hoped the picture was accurate. The CDP promise is that you stop guessing which version of a customer is real. In practice, the hard part was never the software, it was getting every team to agree on what a 'customer' even means.

How it works

A CDP ingests customer data from multiple sources and merges it into one profile per person. The merge uses identity resolution: it decides that a cookie ID, an email address, and a phone number belong to the same individual. The unified profile is then available to other systems in real time, so marketing tools, support desks, and analytics all read the same record. The difference from a data warehouse is the audience layer. A CDP is built so marketers can define a segment and push it to a channel without writing SQL.

Practical uses

Teams use CDPs for cross-channel personalization, identity resolution, and real-time audience activation. A visitor browses the site, the CDP recognizes them from a previous email, and the personalization engine adjusts the next page. Pushing a consistent suppression list to every ad platform is another common job. The contract value shows up as fewer duplicate messages and higher conversion on behavior-based segments.

How to choose

The market split is now obvious. Standalone CDPs like Tealium, Segment, and mParticle charge for compute and identity resolution. Warehouse-native approaches, where the CDP layer runs on top of Snowflake or BigQuery, reuse infrastructure you may already own. Before buying, count your data sources and your identity rules. A small stack with clean keys may not need identity resolution at all, which makes the warehouse-native path dramatically cheaper.

The numbers

Budget reality: standalone CDPs typically run $40,000 to $150,000 per year before usage overages; warehouse-native deployments shift most of that into existing Snowflake or BigQuery spend plus an activation tool. Identity resolution quality varies enough that vendors publish match-rate ranges rather than guarantees - ask for the range on your own sample file before signing anything.

Common mistakes

Buying a CDP to fix data quality is the classic error. The CDP connects to your sources, and if those sources are messy, the profiles inherit the mess. Deduplication at the CDP layer handles some of it, but not inconsistent field naming or missing keys upstream. The second mistake is skipping the governance conversation until after the tool is live, at which point privacy rules become retrofits.

What changed with AI

AI agents need clean, unified profiles to personalize anything. Campaign state, identity, and history all have to live somewhere coherent, and vendors increasingly frame the CDP as the memory layer for agents. The practical effect is that identity resolution changed from a reporting nicety into the foundation for autonomous marketing. If the profile is wrong, the agent personalizes for the wrong person.